Illinois ID photos & facial recognition

The case for it

Efficiency matters. Investigative resources are finite. A system that turns hours or days of manual comparison into a candidate list can free investigators to pursue corroborating evidence and other cases. GAO has recognized speed and objectivity as potential benefits of forensic algorithms.^[9]

The counterargument

Efficiency is not normally the only test for government power. Automation can turn something theoretically possible but labor-intensive into something routine. The policy question is whether the gain in efficiency is proportionate to the additional collection, access, and search capability.

The argument for broader use

Accuracy is not static. Modern algorithms vary enormously in performance, and the best systems can perform far better than older ones under controlled conditions. NIST continuously evaluates current systems and shows substantial differences among algorithms.^[8]

What accuracy does not resolve

Accuracy answers whether the system identifies people correctly. It does not answer whether those people should be identified, which databases should be searchable, how long data should exist, which agencies should have access, or what purposes justify a search.

A perfectly accurate system would eliminate one important objection—misidentification—while making another question sharper: how much identification power should the state possess?

It should be stated carefully. NIST does not support the blanket claim that every facial-recognition algorithm performs poorly on the same demographic groups. Performance varies by algorithm, application, image quality, age, sex, and demographic group; some high-performing algorithms show much smaller differentials than others.^[8]

At the same time, demographic differentials have not simply vanished. NIST's current evaluation continues to measure differences in false-positive and false-negative rates across demographic groups.^[8] That makes procurement, testing, thresholds, image quality, and the consequences of a false candidate relevant to responsible deployment.

NoRec position: privacy objections should not depend on exaggerating algorithmic bias. If disparities improve, that is good. Population-scale biometric search still raises separate questions.

Human review is an important safeguard. The FBI says its facial-recognition results are investigative leads rather than positive identifications and describes trained examiner review and additional investigation before action.^[13] DOJ has likewise recommended trained human review, thresholds, corroboration, and safeguards against automation and confirmation bias.^[12]

But human review is not infallible. A reviewer can be influenced by a ranked candidate list, similarity score, case knowledge, or confirmation bias. Human review also does not resolve collection, retention, access, permissible-purpose, or gallery-scope questions.

Calling a result a lead is a meaningful safeguard when agencies actually enforce that distinction. FBI policy says candidate photos are not positive identification and should not be the sole basis for law-enforcement action.^[13]

A lead can nevertheless affect a real person: who receives scrutiny, whose alibi is examined, who is interviewed, and which evidence investigators seek. DOJ recommends that facial-recognition results not establish probable cause or positive identification without corroboration.^[12]

The argument for use is straightforward: faces are ordinarily exposed in public, and people have always been capable of recognizing one another. Facial recognition automates an observation humans can already make.

The counterargument is that automation changes practical anonymity. A stranger may see your face without knowing your name. A networked system can connect an otherwise anonymous face to an identity at machine speed and repeat that process at scale. GAO has described concerns that widespread facial recognition could enable identification and tracking in public without knowledge or consent.^[10]

There is a legitimate administrative reason for the state to possess a driver's-license or ID photograph: issuing and controlling an identity credential. Illinois law requires the Secretary of State to maintain captured photographs.^[1]

The privacy distinction is secondary use. A photograph collected to issue a credential can also become part of an identification infrastructure useful to agencies pursuing unrelated investigations. Illinois itself treats the images as confidential and specifies who may receive them.^[1]

Confidentiality is a real protection. Illinois statutes and administrative rules limit disclosure, identify eligible recipients, constrain secondary dissemination, and require secure handling in specified circumstances.^[1][7]

“Confidential” does not mean “never accessed.” The same rules enumerate government and criminal-justice uses. The privacy question is the breadth of authorized access and whether the controls are sufficient.

The fact that minors can receive state IDs is not itself evidence of wrongdoing. Children may need identification for ordinary reasons. Illinois says there is no minimum age for a state ID card.^[4]

The relevance is scope. If the public assumes a driver's-license/ID photo system represents only adult motorists, that assumption is incorrect. The credential population can include minors. That matters when evaluating retention, secondary access, and facial-recognition capability associated with those records.

Sharing can prevent duplication and help investigate a case when a useful record is held elsewhere. Illinois administrative rules contemplate dissemination to eligible criminal-justice agencies and impose conditions on secondary dissemination.^[7]

But every additional recipient creates another place where policy, credentials, logging, retention, and security must work. GAO found federal agencies had used non-federal facial-recognition systems without adequate mechanisms to track employee use, creating privacy and risk-assessment problems.^[11]

Nothing about private development automatically makes a system illegitimate. Specialized vendors can build technology that would be expensive for each agency to develop independently, compete on accuracy and cost, and maintain infrastructure.

The concern is incentive and control. A vendor may benefit from expanding customers, searchable data, integrations, or capabilities. Government users can become dependent on systems partly outside public institutions. GAO has documented federal use of non-federal, including commercial, facial-recognition systems and called for improved tracking and risk assessment.^[11]

NoRec position: commercialization is not proof of abuse. It is a reason to ask who owns the infrastructure, who can access the data, what happens when a contract ends, what incentives favor expansion, and which records remain subject to public oversight.

They would solve or reduce several objections. A strong legal predicate can narrow when a search occurs. Audit logs can make misuse detectable. Retention limits reduce historical exposure. Independent review can test compliance. Public policies make rules contestable. Training and corroboration reduce the risk of treating a candidate as an identification.^[12]

Those are meaningful safeguards, not theater.

NoRec's remaining objection is structural: even a well-governed population-scale biometric system creates an identification capability that would not otherwise exist. NoRec favors the narrowest collection and search authority compatible with a demonstrated need. Someone can reasonably conclude that strict safeguards make a particular use acceptable; NoRec can acknowledge that argument while still preferring less surveillance infrastructure.

NoRec's position on this page is narrower: population-scale government biometric search deserves a presumption against expansion and a high burden of justification. Device authentication, one-to-one verification, forensic one-to-many searching, real-time identification, and persistent tracking are materially different applications.

A person can oppose real-time mass identification while accepting a warrant-based forensic search for a violent felony. Another person can reject both. The FAP exists so those distinctions remain visible instead of collapsing the debate into “facial recognition good” versus “facial recognition bad.”